Pearl Uses Bitcoin Mining Power for AI Inference, Miner Revenue Doubles

Author: Castle Labs

Compiled by Deep潮 TechFlow

DeepChao Overview: AI's insatiable demand for computing power is pushing Bitcoin miners toward a precarious crossroads: their electricity and data centers are becoming increasingly valuable, yet the hash rates generated by mining itself are useless. Pearl aims to merge these two functions, enabling miners to run AI inference while maintaining the network. While this narrative has already tripled the token's price, whether actual usage can keep pace is the true determinant of whether it’s “the missing piece AI needs.”

Artificial intelligence is a critical technology. As its usage scales up, demand for computational throughput is rising sharply.

To achieve these throughput levels, AI companies need GPUs and energy. However, even well-funded companies may struggle to secure sufficient resources due to limited GPU supply and scarce electricity.

A common solution is to lease infrastructure.

Over the past few years, Bitcoin miners have built industrial-scale mining infrastructure and secured the most critical resource: access to electricity. As demand for AI computing power grows, this advantage is being revalued by the market. Although these miners operate ASICs, which cannot be directly used for AI calculations, miners can lease their facilities and electricity to AI clients.

As a result, miners have the opportunity to generate a more sustainable revenue stream by using electricity for AI-related computations. The revenue structure of mining companies already reflects this shift: among 11 publicly traded mining firms, the proportion of AI and HPC revenue to total revenue rose from 2% in the first quarter of 2025 to 22% in the second quarter of 2026.

As a result, Bitcoin hash power may face increased competition from AI inference.

Another way to observe the growing relationship between AI and mining infrastructure is to examine how BTC is mined. Under the Proof-of-Work mechanism, miners deploy vast amounts of computational power to secure the Bitcoin network and validate transactions. This process requires continuous calculation of cryptographic hashes, which have no inherent value once computed. Mining is economically viable only when the value of the mined BTC exceeds the costs of computation and infrastructure.

Then why can't you have both?

Computations can simultaneously run AI inference while continuing to secure the network and mine assets; this approach is known as Proof-of-Useful Work (PoUW). The Pearl network, which we will discuss in detail in the next section, focuses on the intersection of AI adoption and blockchain.

Pearl is an L1 blockchain that launched in April this year, forked from Bitcoin’s code but replacing SHA-256 hashing with matrix multiplication (matmul)—the type of operation GPUs perform during AI inference.

This is exactly what Pearl is doing: transforming Bitcoin’s proof of work into useful computation for AI (PoUW).

To integrate AI inference into mining, Pearl designed a system that replaces Bitcoin’s standard SHA-256 hash verification with verification of matrix multiplication.

Matrix multiplication is how AI training and inference are performed, and it requires significant computational power.

Matrix multiplication and SHA-256 share many similarities:

Fast solving requires extensive computation.

Algorithm results are not subjective judgments.

The answer cannot be forged or guessed (you also can't predict in advance which function will be used for the next block).

Accordingly, these computations must still be “expensive enough to perform, but the answers must be objective and cheap enough for others to verify.” This is precisely why Pearl uses zero-knowledge (ZK) proofs—to make verification inexpensive.

To meet AI-related compute demands, Pearl partners with aggregated compute networks and inference service providers like Together AI. This enables enterprises to utilize compute power generated by miners, while miners earn PRL tokens. Additionally, Together AI operates mining infrastructure for the network and offers discounts to its customers using the PRL rewards it receives.

Ultimately, Pearl aims to replace the computations performed by BTC miners for mining Bitcoin with more useful matrix multiplications commonly used in AI, while keeping the core blockchain economic model intact and prioritizing AI.

However, the “useful” aspect of PoUW deserves further discussion. One criticism of Pearl is that the protocol cannot prove these matrix multiplications actually come from paying customers or useful work, as miners could input synthetic matrices and still receive PRL rewards.

In fact, a paper from June found that the prevailing mining software at the time produced results that were “verifiable matrix multiplications, but with no external reasoning output.” Brian Breslow also noted this in his PRL investment thesis.

This is precisely the key: avoid making subjective judgments about the "usefulness" of reasoning. Bittensor has previously implemented a similar system:

Scarce native token

Miners provide AI-related resources.

Validators assess output quality

Rewards flow to areas the network considers valuable.

However, this design introduces subjective judgment and greater discretion. There have been previous instances of collusion, as "validators have an incentive to collude and vote on usefulness."

In contrast, Pearl's AI reasoning involves no subjective judgment.

The most critical point at this level is that the entire value proposition rests on the assumption that submitting two matrices for AI inference is simple and incurs almost no additional overhead—the value received from PRL must exceed the integration friction, risk, throughput loss, and costs.

Miners must still cover the full cost of their GPUs to earn PRL. As difficulty increases, only miners with top-tier GPUs can remain profitable, as they possess the strongest mining power.

If enough "useful miners" join the network, others will no longer profit and will have no incentive to continue mining.

Avoiding any factors that could introduce subjectivity helps prevent miner collusion issues that have occurred in other networks, such as Bittensor.

However, there is a prerequisite: currently, due to the network using a simplified digital format (INT8), the additional overhead is approximately 10–15%. Only when this overhead is reduced sufficiently can useful miners become the most efficient block producers. Efforts to address this issue are already underway: Pearl-GEMM has been deployed on the testnet, where performance loss is reduced to just 0–3% when using standard formats.

The PRL token has a hard cap of 2.1 billion. Unlike Bitcoin’s halving mechanism, PRL’s release follows a linear decay schedule, with approximately 50% of the total supply expected to be released by the fourth year.

As of the time of writing, the market capitalization of PRL is $475 million, with 339 million PRL already mined (16.2% of the circulating supply). An additional 305 million PRL are expected to be issued over the next 12 months, nearly doubling the circulating supply.

PRL is currently listed on two exchanges (BigONE and SafeTrade), with a daily trading volume of approximately $6 million and a buy order depth of around $22,000 within a 2% price range.

Nevertheless, PRL's repricing appears to have outpaced actual usage. Since September, PRL has more than tripled in value, gaining the most attention after Brian Breslow of No Limit Holdings presented his investment thesis.

In addition, PRL is currently in a state of dilution. Pearl has no halving mechanism, and approximately 1.01 million PRL are issued daily to miners, valued at around $1.4 million as of this writing. More critically, the PRL supply did not follow its intended distribution schedule: during the first six days, when mining difficulty was lowest, about 122 million PRL (36% of the current circulating supply) were issued, and 20,000 blocks were mined within the first 30 hours after launch. Since then, block times have stabilized near the target of 194 seconds. This early lead caused unnecessary dilution, benefiting miners who participated in the first week and exacerbating supply concentration.

The token itself does not generate any income. Miners mine PRL and earn stablecoins by providing AI inference services, but PRL has no burn mechanism and its supply will not decrease.

However, the team is actively addressing current issues, including performance overhead, synthetic mining, and token value accumulation:

Pearl-GEMM Upgrade: The current version of Pearl only handles simplified numeric formats (INT8), which most compatible AI models do not use, resulting in a 10–15% reduction in model speed. Under this overhead, AI companies have little incentive to purchase compute power from miners. However, a solution is already being tested on the testnet: Pearl-GEMM can process standard formats with only a 0–3% speed loss. This upgrade is critical, as AI operators will not accept a 10–15% reduction in output.

Real AI work must replace synthetic mining: Not all mining currently involves useful AI inference. A June study examined approximately 8,000 machines running on hardware capable of inference but executing no AI-related code. Pearl’s envisioned decentralized compute market can address this issue by partnering with AI companies: users can pay and earn rewards in PRL to run inference on the network. This includes reserving GPU capacity, fine-tuning models, or settling compute contracts without relying on traditional cloud solutions.

AI creates buying pressure and a potential burn mechanism: Currently, the use of AI inference generates selling pressure. U.S. cloud provider Together AI launched a model in May 2026 and offered a 25% discount for access, with the discounted portion earned through PRL mining to meet computational demands. The team is considering pricing AI services in PRL or burning a portion of PRL with each purchase, aiming to move away from USD-based pricing and generate additional demand for the asset.

Pearl is making computing power useful, and given the growth and eventual widespread adoption of AI, its target market is highly attractive.

Building a new network always comes with challenges, and for Pearl, these include high costs, synthetic mining, and token value accumulation mechanisms—all of which the team is addressing one by one.

Most importantly, Pearl currently prices its services in USD, with little value flowing back to the token; using PRL would be fundamental to driving token value and improving demand accumulation. The majority of minted PRL is used as incentives and may ultimately be sold by miners for profit.

Additionally, PRL is expected to serve as the utility token for the decentralized market envisioned by the team, making traffic and trading volume difficult to predict. At this stage, we believe directly comparing PRL’s price to BTC (e.g., Pearl reaching 1% of BTC’s market cap) is more speculative than factual.

However, we believe this presents a compelling value proposition for BTC miners already selling AI inference, with the potential to capture a significant portion of mining hash rate, especially given the future growth of the AI inference market.